ResearchBenchmarksMultiple Choice Question Answering on Include cFollow30.43Normalized AccuracyML26.665227.642628.6229.5974Apr 22, 2026Evaluation ResultsMethodMethodLinksNormalized AccuracyML2026.0430.43No Filtering2026.0430.25MKC-e2026.0429.53HQ seedVariant=SeedVariant=Seed2026.0428.44ML (10%)Retention=10%Retention=10%2026.0428.44ML (20%)Retention=20%Retention=20%2026.0427.9HQ (20%)Retention=20%Retention=20%2026.0427.72ML (Q3)Sampling Strategy=Q3Sampling Strategy=Q32026.0427.54HQ (10%)Retention=10%Retention=10%2026.0427.17HQ2026.0426.81